Stability and overclocking - what do you consider stable

thecold

[H]ard|Gawd
Joined
Nov 12, 2017
Messages
1,216
I've seen various definitions of stability.

Some people will call prime95 11 minutes stable. Others will run a few Cinebench r20's and call it good. This a screenshot of my pc running an antivirus full system scan on my intel p3700. Notice the max temperature. It's actually higher than some AVX2 Blender loads on my desktop (bmw in this example). This blender load generally sits at 72-74c after heavy pre-loading.

Cinebench is a bit higher at about 2 degree's higher. This is after looping it for about 20 minutes (this was for temperature, not for stability).

My personal opinion is that if it's not stable with multiple stress tests for long periods of time, It's not stable.

What do you consider stable?

Before I take stability overclock screenshots seriously, it needs to be at least a couple hours.
 

Attachments

  • blendervsdefender.JPG
    blendervsdefender.JPG
    130.9 KB · Views: 0
No errrors in RAM tests. I do two RAM tests. One before windows boots, which usually returns problems within 30 seconds. and one while windows is running, which I run over night.

20 passes of prime 95 small FFT

completes a 5+ minute handbrake job without crashing/freezing

and then after that......basically never randomnly crashing from whatever I might be doing. It can still happen, despite stress tests.
 
Last edited:
~5hrs OCCT and I consider it good to go. Then load it with WCG BOINC WUs and check it for returned errors for a week or so. Then it's rock solid.
 
Intel burn test on very high or maximum for me @ 5 to 10 passes without issue is my personal indicator.
 
11 minutes stable? I don't know many computers that can encode high quality h265 in 11 minutes. Stables means it NEVER crashes due to the overclock. Granted, to truly achieve this you'd need to run your stress test infinitely. Obviously a compromise is needed.

Myself, I run my stress test overnight and once I have the settings I think are final I'll usually run for 24 hours. At that point, the chances of crashing due to overclock is almost nothing.

I have often found stability issues in these marathon tests. Even after 3 or 4 hours. Usually all it takes is a very minor tweak to correct. I'm also not interested in bragging about how high I was able to push it. I'm in it purely for the free performance boost. 50Mhz or less isn't going to be missed.
 
  • Like
Reactions: ep3w
like this
Your computer is nothing more than a powerful calculator. When you run an unstable oc, it goes from being a calculator to a random number generator. Make sure you make good numbers lol.
 
I've been thinking about this while looking for optimal memory speed and timings. Basically, if Firefox doesn't crash with a bunch of open tabs and I get no CRC errors while extracting files, I move on to the first few MemTest tests. If it passes those it's probably stable enough for anything I'll ever do.
 
For me it's not a particular stress test. Usually I will do many of them but the gold standard is uptime. Can I do all that I normally do and not restart for a month? If so I consider it good.
 
I haven't pushed my new build yet but on my old one I've had prime95 detect errors after 18 hours.
 
I ran prime95 for five minutes and it's good. Gets pretty boring watching the system thermally throttle down. As long as I can see that, there's little point in continuing.
 
Last edited:
Post? Check.
Boot to desktop? Check.
Finish SuperPi 1M? Check.

Stable enough in my book.
 
I can't get 2 different RAM diagnostics to find bad bits in 24 hours of testing with each, at a computer case temperature of 40 Celcius or with the RAM heatsink or chip packages at 50-60C.
 
MemTestx86+ with test 5 and 8 for 2 passes each (to dial in settings)
9 hours memory testing, all tests (after settings are perceived stable)
Cinebench 3 hours (Furmark+Prime95 concurrently for 4 hours for older systems)
48 hours Prime95 blend

If the system is going to be contributing to a distributed project, I also throw in a weekend of Linpack just to be sure. Every 3 months, when I update my monitor calibrations, I usually run Cinebench again overnight to determine if the system is still operating within spec.
 
I ran prime95 for five minutes and it's good. Gets pretty boring watching the system thermally throttle down. As long as I can see that, there's little point in continuing.

You know... the first 5-10 minutes of P95 is just self testing.
 
  • Like
Reactions: tived
like this
MemTestx86+ with test 5 and 8 for 2 passes each (to dial in settings)
9 hours memory testing, all tests (after settings are perceived stable)
Cinebench 3 hours (Furmark+Prime95 concurrently for 4 hours for older systems)
48 hours Prime95 blend

If the system is going to be contributing to a distributed project, I also throw in a weekend of Linpack just to be sure. Every 3 months, when I update my monitor calibrations, I usually run Cinebench again overnight to determine if the system is still operating within spec.

Your stability tests are awesome. That's all I have to say.
 
so what. It's testing, loading the cpu and checking for errors.

Huh? I don't get your attitude. It's a self test, it hasn't even begun to load the cpu. Now not sure your stance since you don't seem to give a shit either way with a 5 minute run. Point is, it hasn't even begun load testing.
 
Huh? I don't get your attitude. It's a self test, it hasn't even begun to load the cpu. Now not sure your stance since you don't seem to give a shit either way with a 5 minute run. Point is, it hasn't even begun load testing.
Don't act like a know-it-all and you will get less attitude. I can read a cpu utilization graph along with the cpu temps. It's loading the CPU.self-test or not.
 
Last edited:
You have some issues man.
My issue is people stating incorrect things and not even taking one minute to look for themselves at what is going on.

is the cpu loaded? YES

is the cpu heating up? YES

is the os/computer/prime95 stable? YES

Self-test or not - DOESN'T MATTER.

Empirical evidence suggests this is several ways. What evidence do you have that the self-test mode does NOT constitute a load test on the cpu? As far as I can tell, you read "self-test" and made a bad assumption - by being ignorant of the system status.
 
Last edited:
My issue is people stating incorrect things and not even taking one minute to look for themselves at what is going on.

Ugh, you're full of shit if you ar4e telling us 5 minutes of prime proves stability. Like I wrote, you're just trolling now cuz 5 minutes of p95 doesn't prove jack shit.
 
Ugh, you're full of shit if you ar4e telling us 5 minutes of prime proves stability. Like I wrote, you're just trolling now cuz 5 minutes of p95 doesn't prove jack shit.

I'm telling you that you stating "test-mode" does not stress the cpu and you are full of shit. Stop changing your argument from "test-mode doesn't load the cpu" to "a five minute test doesn't prove stability". Stability testing is obviously subjective. I am not even going there. Load test the way you want, I don't care.

Stating the P95 self-test doesn't load the system is a misinformed lie.

LOOKS LOADED TO ME

upload_2020-1-15_17-46-11.png


upload_2020-1-15_17-46-30.png
 
Last edited:
>100 hours of GTA V, Witcher3, ME Andromeda, Prime95, AIDA 64, OCCT - rock stable.
PUBG - BSOD.
???
I don't know what to believe anymore.
 
>100 hours of GTA V, Witcher3, ME Andromeda, Prime95, AIDA 64, OCCT - rock stable.
PUBG - BSOD.
???
I don't know what to believe anymore.
Believe this - pugb is a very poorly coded game. Do you remember the cause of the BSOD? Page Fault?
 
Believe this - pugb is a very poorly coded game. Do you remember the cause of the BSOD? Page Fault?

Haha I was going to post the same thing, those devs are a complete joke. However it has been a while since I played it, when I did last (early 2019) it did not crash nor crash the entire system.
 
2 kinds of stable.

1. OC to beat a Record- prime 95 for a few minutes maybe 1 pass with memtest86+ or skip this altogether and just start benchmarking.
2. OC to use long term- prime 95 5+hours, 4-8 passes with memtest86+ 0 errors at the very least.
 
24 h Prime95
8 h RealBench stress + 10 benchmark passes
>400% memtest
real life usage
 
I've seen prime fail after 24 hours, or have had issues regardless of it passing, so to me, prime95 is a complete waste of time. I haven't used memtest in at least a decade :D

I used to use IntelBurnTest, now I run Linpack Xtreme for testing cpu and mems for 24/7 oc. If not for 24/7 if it passes the bench with a better score than before its a win.
 
For quick check if my memory clocks and timings are even remotely viable, 1 pass Memtest86. For checking if it is actually stable, overnight of HCI Memtest through the DRAM calculator, something like 2000% for all threads.

For CPU it is 12 hours of latest version Prime95, Small FFTs with AVX. And another 12 hours of Prime95 Large FFT to add some memory and memory controller stability testing in the mix. That said, now that I am on Ryzen 2 this is all prettty pointless because you cannot overclock the damn things in any meaningful way. I just enabled PBO and clock override, applied a mild 0.050v undervolt offset to keep the runaway volts and temps in check and CPU handles the rest. Booooring....

Anyway, I OC my rig to use it. I don't want my files get corrupted just because my ram is falling apart or my CPU, as someone pointed out, becomes a "random number generator" in a wrong place at wrong time.
 
300 seconds of r20. Overwatch runs like a dream no crashes. YouTube Facebook works just as well. Yup I’m happy. Can’t wait for Nvidias next gen x80 cards to replace my 1080 FE
 
Last edited:
At least one hour of Prime95, 20 pases of Lynx and Heaven Benchmark to test CPU, RAM and GPU stability.
 
48 hours of prime 95 mixed for the cpu.

24-48 hours of Unigine or something demanding on a loop for GPU.

I have seen systems be stable for 36 hours and then fail just after.

There is no substitute for thorough LONG stability testing.

It may seem long, but once you arrive at your final settings you only have to do it once, and then it is done.
 
Last edited:
Bumping because I decided to update my bios and lower my overclock this weekend and test stability.
This has been my method for around 10 years and I never had a random crash from instability. Before RealBench came out I used SiSoftSandra stress in its place.

For CPU:
5 Runs of Cinebench
20 Loops of 3DMark Or TimeSpy CPU Test Only
10 Loops of Intel Burn Test (High) (usually by this point I would know if the system is stable enough to continue the longer tests)
10 Loops of Intel Burn Test (16 GB)
1 Hour / Prime 95 12/12 AVX OFF
30 Min ROG Real Bench (32GB)
2 Hours Prime 95 Blend

Entire process takes like 4-5 hours and I use the machine during the Prime 95 testing. I think anything longer is a waste of time.

For GPU:
Unigine Superposistion (For Initial Testing Overclock and Memory)
TimeSpy Stress Test (takes about 15mins)
Furmark 1 Hour
Gaming
 
Last edited:
Back
Top